Dynamic path recommendation method and device, vehicle machine system and storage medium

Abstract

The application provides a dynamic path recommendation method and device, a vehicle system and a storage medium. The method comprises the following steps: extracting the time and position of a trajectory data point in a historical driving trajectory of a vehicle, and generating a time domain vector of an interest point for the point; extracting the destination type, position and arrival time in a historical navigation record of a user, and generating a space domain vector of the interest point; matching the space domain vector of the interest point with the time domain vector of the interest point for the same or similar points, obtaining the length of the stay time of the interest point, and generating a space-time feature map of the user behavior habit; sequentially performing multi-layer space-time convolution processing on the space-time feature map of the user behavior habit, wherein the features obtained by the previous layer of space-time convolution network processing are input into the next layer of space-time convolution network after normalization processing, and a heat map of the user behavior habit is obtained; and determining a recommended path according to the current position of the vehicle, the current real-time road condition and the heat map of the user behavior habit. The application can obtain a customized dynamic path recommendation in line with the user behavior habit.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application generally relate to the field of vehicle navigation, and more particularly, to a dynamic path recommendation method, device, car machine system and storage medium. BACKGROUND

[0002] With the rapid development of domestic economy and the increase of urban resident density, the number of urban vehicles is increasing, and the urban traffic problem is becoming increasingly serious.

[0003] Having an efficient dynamic path induction algorithm is one of the effective ways to improve road resource utilization and traffic speed and to relieve congestion. However, according to research results at home and abroad, most of the existing dynamic path induction algorithms only perform dynamic path calculation according to real-time road conditions, and only take road condition information as a prerequisite for path calculation, without considering the behavior habits of travelers, which limits the selection space of users. When encountering congestion that cannot be avoided, no other solutions can be provided, which brings a poor driving experience to the vehicle owner and cannot bring satisfactory emotional value to the vehicle owner. SUMMARY

[0004] In order to solve the above problems in the prior art, in a first aspect, embodiments of the present application provide a dynamic path recommendation method, the method comprising: obtaining a vehicle historical driving trajectory and a user historical navigation record; extracting the time and position of a trajectory data point in the vehicle historical driving trajectory; generating a point of interest time domain vector according to the time and position of the trajectory data point; extracting the destination type, destination position and arrival time of each record in the user historical navigation record; generating a point of interest space vector according to the extracted destination type, destination position and arrival time in the user historical navigation record; matching each point of interest in the point of interest space vector with a point of interest in the point of interest time domain vector that is the same or adjacent in position to obtain the length of time spent at each point of interest; generating a user behavior habit space-time feature map according to the length of time spent at each point of interest, the point of interest position and the point of interest type; converting the user behavior habit space-time feature map into a graph data format; sequentially performing multiple times of space-time convolution processing on the user behavior habit space-time feature map in the graph data format through a multi-layer space-time convolution network, wherein each layer of the multi-layer space-time convolution network includes a time series graph convolution network and a spatial graph convolution network, and wherein the features obtained by the previous layer of space-time convolution network are input into the next layer of space-time convolution network after normalization processing; performing normalization processing and up-sampling on the features obtained by the multi-layer space-time convolution network to obtain a user behavior habit heat map; determining a recommended path according to the current position of the vehicle, the current real-time road condition and the user behavior habit heat map.

[0005] In some embodiments, after the user behavior habit space-time feature map is generated, the method further comprises: assigning a weight to each interest point through an attention mechanism according to the priority of each interest point; and updating the user behavior habit space-time feature map according to the weight assigned to each interest point.

[0006] In some embodiments, the method further comprises: receiving an operation of the user on the recommended path; and updating the weight of the interest point that is the destination of the recommended path according to the operation of the user.

[0007] In some embodiments, performing the multiple space-time convolution processes comprises: inputting the unnormalized features output by the first layer of the multi-layer space-time convolution network to each subsequent layer of the space-time convolution network.

[0008] In some embodiments, performing the multiple space-time convolution processes comprises: in each layer of the space-time convolution network, first processing using a time graph convolution network and then processing using a space graph convolution network.

[0009] In some embodiments, the method further comprises: after all the interest points in the interest point space vector have been matched, if there are still interest points that have long-term stays and have not been matched in the interest point time vector, extracting the positions, arrival times and stay times of these interest points; determining the type of the interest points according to the positions of the interest points; and filling the extracted positions, arrival times, stay times and types of the interest points into the user behavior habit space-time feature map.

[0010] In some embodiments, the multi-layer space-time convolution network comprises three layers of space-time convolution networks.

[0011] In some embodiments, the method further comprises: sending a prompt message through a human-computer interface to ask whether to update the model; receiving a selection of the prompt message by the user; and in the case where the user selects to update the model, performing model upgrading on the multi-layer space-time convolution network using the recent user behavior habit space-time feature map.

[0012] In a second aspect, embodiments of the present application provide a dynamic path recommendation device, comprising: a historical data acquisition module configured to acquire a vehicle historical driving trajectory and a user historical navigation record; a trajectory data extraction module configured to extract time and position of a trajectory data point in the vehicle historical driving trajectory; a time-domain vector generation module configured to generate a point of interest time-domain vector according to the time and position of the trajectory data point; a navigation data extraction module configured to extract a destination type, a destination position and an arrival time of each record in the user historical navigation record; a space-domain vector generation module configured to generate a point of interest space-domain vector according to the extracted destination type, destination position and arrival time in the user historical navigation record; a matching module configured to match each point of interest in the point of interest space-domain vector with a point of interest in the point of interest time-domain vector that is the same or adjacent in position, to obtain a length of time of staying at each point of interest; a space-time feature map generation module configured to generate a user behavior habit space-time feature map according to the length of time of staying at each point of interest, a point of interest position and a point of interest type; a format conversion module configured to convert the user behavior habit space-time feature map into a graph data format; a space-time convolution module configured to sequentially perform multiple space-time convolution processing on the user behavior habit space-time feature map in the graph data format through a multi-layer space-time convolution network, wherein each layer of the multi-layer space-time convolution network comprises a time series graph convolution network and a spatial graph convolution network, and wherein a feature obtained by a previous layer of space-time convolution network processing is input into a next layer of space-time convolution network after normalization processing; a heat map acquisition module configured to normalize and up-sample a feature obtained by the multi-layer space-time convolution network to obtain a user behavior habit heat map; and a path determination module configured to determine a recommended path according to a current position of a vehicle, a current real-time road condition and the user behavior habit heat map.

[0013] In some embodiments, the device further comprises an attention mechanism module configured to: after generating the user behavior habit space-time feature map, assign a weight to each point of interest through an attention mechanism according to a priority of each point of interest; and update the user behavior habit space-time feature map according to the weight assigned to each point of interest.

[0014] In some embodiments, the device further comprises a weight update module configured to: receive an operation of a user on the recommended path; and update a weight of a point of interest that is a destination of the recommended path according to the operation of the user.

[0015] In some embodiments, the space-time convolution module is further configured to: input an unnormalized feature output by a first layer of space-time convolution network in the multi-layer space-time convolution network into each subsequent layer of space-time convolution network.

[0016] In a third aspect, embodiments of the present application provide a car machine system, comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, implements the dynamic path recommendation method described in any of the above embodiments.

[0017] In a fourth aspect, embodiments of the present application provide a storage medium storing computer readable instructions which, when executed by a processor, perform the dynamic path recommendation method described in any of the above embodiments.

[0018] Embodiments of the present application provide a dynamic path induction scheme which not only recommends a path based on real-time traffic conditions in the city, but also guides the user to a suitable point of interest for corresponding activities to wait for the congestion state to be alleviated or ended according to the user's behavior habits when congestion occurs.

[0019] The technical solution adopted by embodiments of the present application to solve the technical problem is: a city point of interest heat map is established based on user behavior habits. In order to establish the point of interest heat map, the vehicle driving trajectory and the user navigation record need to be obtained first, then the two kinds of data are fused to obtain a user behavior habit space-time feature map, then a graph neural network model is established for the data, and finally the user behavior habit space-time feature map is input into the model for training to obtain a city point of interest heat map for guiding the user.

[0020] Whenever the user travels and encounters an unavoidable congestion state, the car machine system will make a congestion avoidance recommendation for the user according to the user's city point of interest heat map, guide the user to a point of interest that meets his behavior habits, and improve the user experience and satisfaction. BRIEF DESCRIPTION OF DRAWINGS

[0021] The above and other objects, features and advantages of embodiments of the present application will be readily understood through reading the detailed description of embodiments of the present application below, with reference to the accompanying drawings. In the drawings, several embodiments of the present application are shown by way of example and not limitation, in which:

[0022] Figure 1 A flowchart of a dynamic path recommendation method according to an embodiment of the present application is shown;

[0023] Figure 2 A schematic diagram of generating a space-time feature map according to an embodiment of the present application is shown;

[0024] Figure 3 A schematic diagram of a first example of space-time convolution processing according to an embodiment of the present application is shown;

[0025] Figure 4 A schematic diagram of a second example of space-time convolution processing according to an embodiment of the present application is shown;

[0026] Figure 5 FIG. 6 shows a schematic diagram illustrating a third example of spatio-temporal convolution processing according to an embodiment of the present application;

[0027] Figure 6 FIG. 7 shows a block diagram of a dynamic path recommendation device according to an embodiment of the present application.

[0028] In the drawings, identical or corresponding reference signs indicate identical or corresponding parts. DETAILED DESCRIPTION

[0029] The principles and spirits of the present application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are given only so that those skilled in the art can better understand and implement the present application, and do not limit the scope of the present application in any way.

[0030] In one aspect, embodiments of the present application provide a dynamic path recommendation method. Referring to FIG. 1, a flowchart of a dynamic path recommendation method 100 according to an embodiment of the present application is shown. As an example, the method can be executed on a car machine system of a vehicle, or can be executed on a server (e.g., a cloud server), or part of the steps are executed on the car machine system and part of the steps are executed on the server, and the present application does not limit in this aspect. Figure 1 As shown in FIG. 1, the method 100 includes steps S101-S111.

[0031] Figure 1 As shown in FIG. 1, the method 100 includes steps S101-S111.

[0032] In step S101, a vehicle historical driving trajectory and a user historical navigation record are obtained.

[0033] In step S102, time and location (e.g., latitude and longitude) of trajectory data points in the vehicle historical driving trajectory are extracted.

[0034] In step S103, a point of interest time domain vector is generated according to the time and location of the trajectory data points.

[0035] In step S104, destination type, destination location and arrival time of each record in the user historical navigation record are extracted.

[0036] In step S105, a point of interest space domain vector is generated according to the extracted destination type, destination location and arrival time in the user historical navigation record.

[0037] In step S106, each point of interest in the point of interest space domain vector is matched with a point of interest in the point of interest time domain vector with the same or adjacent location, and a length of time of staying at each point of interest is obtained.

[0038] ​In step S107, a user behavior habit space-time feature map is generated according to the length of stay at each point of interest, the point of interest location, and the point of interest type.

[0039] Reference Figure 2 which shows a schematic diagram of generating a space-time feature map according to an embodiment of the present application. Through the above steps, in the process of generating the user behavior habit space-time feature map, the vehicle driving trajectory and the user navigation record are data fused. In order to obtain the fused data, first, the vehicle driving trajectory and the user navigation record are data refined. For the vehicle driving trajectory, the time and the latitude and longitude of the trajectory data point need to be extracted, and a point of interest time domain vector is established for the latitude and longitude. The vector is used to record the location of the vehicle at each time point in the time dimension. For each navigation record of the user, the latitude and longitude of the destination, the type of the destination, such as a restaurant, a shopping mall, etc., which is converted into a corresponding enumeration value for easy calculation, and the arrival time of the destination are extracted. After all the navigation records are data extracted, a point of interest space domain vector is obtained, which is used to record the points of interest visited by the user. When the point of interest time domain vector and the point of interest space domain vector are obtained, each point of interest in the space domain vector is matched with the point in the time domain vector that has the same latitude and longitude or is within a similar range, and the time of stay of the user at each point of interest is obtained. In this way, the space-time feature map representing the user behavior habit is obtained.

[0040] As an embodiment of the present application, the method can further include: after all the points of interest in the point of interest space domain vector have been matched, if there are still long-stay points of interest in the point of interest time domain vector that have not been matched, extracting the location (such as latitude and longitude), arrival time, and stay time of these points of interest; determining the type of the point of interest according to the location of the point of interest; filling the extracted location, arrival time, stay time, and type of the point of interest into the user behavior habit space-time feature map, thereby obtaining a complete user behavior habit space-time feature map.

[0041] In step S108, the user behavior habit space-time feature map is converted into a graph data format.

[0042] In step S109, the user behavior habit space-time feature map in the graph data format is sequentially subjected to multiple times of space-time convolution processing by a multi-layer space-time convolution network. Each layer of the multi-layer space-time convolution network includes a time series graph convolution network and a spatial graph convolution network, and the feature obtained by the previous layer of space-time convolution network is input into the next layer of space-time convolution network after being normalized. The multi-layer space-time convolution network and the normalization processing therebetween constitute a graph neural network model for analyzing the user behavior habit space-time feature map. As an embodiment of the present application, the multi-layer space-time convolution network can be a three-layer space-time convolution network.

[0043] Since the user behavior habit space-time feature map belongs to non-Euclidean space data, a conventional convolutional neural network cannot well process such data, and therefore, embodiments of the present application adopt a graph convolutional neural network to analyze and process user behavior habit feature data, and meanwhile, due to the space-time characteristics of the behavior habit data, the embodiments of the present application adopt a space-time convolution manner to extract features, and finally obtain a probability distribution map of a user appearing at a certain interest point at a certain time, that is, a user behavior habit heat map.

[0044] Reference Figure 3 which shows a schematic diagram of a first example of space-time convolution processing according to an embodiment of the present application.

[0045] In this first example, first, the user behavior habit space-time feature map is converted into a format of graph data to facilitate the calculation of the graph neural convolution network. Optionally, data enhancement can also be performed. Subsequently, the data is sent into a continuous three-layer space-time convolution network model for training, the time series convolution is used to extract time series features, and the spatial convolution is used to extract spatial features. As an embodiment of the present application, performing multiple space-time convolution processing includes: in each layer of the space-time convolution network, first, processing is performed by using a time series graph convolution network, and then processing is performed by using a spatial graph convolution network. After each space-time convolution, the output data is normalized to ensure that the data distribution conforms to the normal distribution as much as possible and to ensure the rationality of the output result, and finally, the final data is up-sampled to obtain a user behavior habit heat map.

[0046] In step S110, the features obtained by the multi-layer space-time convolution network are normalized and up-sampled to obtain a user behavior habit heat map.

[0047] In step S111, a recommended path is determined according to the current position of the vehicle, the current real-time road condition and the user behavior habit heat map. When the user encounters congestion, the car machine system will combine the real-time road condition and the user behavior habit heat map to provide the user with more personalized and customized path induction services.

[0048] The dynamic path recommendation method proposed by the embodiments of the present application is not limited to real-time road condition information when performing path induction, but also utilizes user behavior habit data to provide the user with customized services with higher emotional value.

[0049] As an embodiment of the present application, after generating the user behavior habit space-time feature map, the method can further include: assigning a weight to each interest point by using an attention mechanism according to the priority of each interest point; and updating the user behavior habit space-time feature map according to the weight assigned to each interest point.

[0050] Reference Figure 4FIG. 6 shows a schematic diagram illustrating a second example of the spatio-temporal convolution processing according to an embodiment of the present application. In the second example, the user behavior habit spatio-temporal feature map is first converted into the format of graph data to facilitate the calculation of the graph neural convolution network. Optionally, data augmentation can also be performed. If there are many points of interest, a corresponding priority can be assigned to each point of interest, and an attention mechanism can be adopted to distinguish each point of interest and give different weights. The priority can be determined according to one or more of the following factors: the point of interest stay time, the point of interest type or attribute, the number of visits, the point of interest location, the distance from home or company, etc. The data is then fed into a continuous three-layer spatio-temporal convolution network model for training, in which the temporal convolution is used to extract temporal features and the spatial convolution is used to extract spatial features. As an embodiment of the present application, the multiple spatio-temporal convolution processing includes: in each layer of the spatio-temporal convolution network, first processing by the temporal graph convolution network, and then processing by the spatial graph convolution network. The output data is normalized after each spatio-temporal convolution to ensure that the data distribution conforms to the normal distribution as much as possible, and the rationality of the output result is ensured. Finally, the final data is up-sampled to obtain the user behavior habit heat map.

[0051] In the embodiment in which the attention mechanism is introduced, as a further embodiment, the method can further include: receiving the user's operation on the recommended path; and updating the weight of the point of interest as the destination of the recommended path according to the user's operation. After obtaining the user's permission, the weight of the point of interest can be adjusted according to the user's selection of the recommended path after the path recommendation is made to the user, and the weight of the point of interest will affect the generation of the heat map in the attention mechanism, so as to timely adjust the recommendation strategy and ensure that the strategy has a certain timeliness, thereby bringing a better user experience.

[0052] As an embodiment of the present application, the multiple spatio-temporal convolution processing includes: inputting the un-normalized features output by the first layer of the spatio-temporal convolution network in the multi-layer spatio-temporal convolution network to each subsequent layer of the spatio-temporal convolution network.

[0053] Reference Figure 5This diagram illustrates a third example of spatiotemporal convolution processing according to an embodiment of the present invention. In this third example, the spatiotemporal feature map of user behavior habits is first converted into graph data format to facilitate computation by the graph neural convolutional network. Optionally, data augmentation can also be performed. The data is then fed into a three-layer spatiotemporal convolutional network model for training, with temporal convolution used to extract temporal features and spatial convolution used to extract spatial features. As an embodiment of the present invention, performing multiple spatiotemporal convolution processes includes: in each layer of the spatiotemporal convolutional network, processing is first performed using a temporal graph convolutional network, followed by processing using a spatial graph convolutional network. After each spatiotemporal convolution, the output data is normalized to ensure that the data distribution conforms to a normal distribution as much as possible, ensuring the rationality of the output results. Finally, the final data is upsampled to obtain a user behavior habit heatmap. Furthermore, considering the sparsity of feature data, the unnormalized features output by the first layer of the multi-layer spatiotemporal convolutional network can be input into each subsequent layer of the spatiotemporal convolutional network, i.e., a residual structure is introduced. Figure 5 As shown, the unnormalized output of the first spatiotemporal convolutional network is input into the second and third spatiotemporal convolutional networks. Thus, the second spatiotemporal convolutional network receives both the unnormalized and normalized outputs of the first layer as input, while the third layer receives both the unnormalized output of the first layer and the normalized output of the second layer. In this example, by introducing a residual structure into the three spatiotemporal convolutional layers, gradient explosion or vanishing gradients are avoided, improving the model's robustness.

[0054] As one embodiment of the present invention, the method may further include: sending a prompt message on whether to update the model through a human-computer interface; receiving the user's selection of the prompt message; and, if the user selects to update the model, upgrading the multi-layer spatiotemporal convolutional network using spatiotemporal feature maps of recent user behavior habits.

[0055] For example, the vehicle's infotainment system can periodically ask users if they want to update the model. After obtaining the user's consent, the system can upload recent spatiotemporal feature maps of user behavior to the cloud for model fine-tuning and online model upgrades, providing users with a better user experience and meeting their ever-changing needs.

[0056] In another aspect, embodiments of the present invention provide a dynamic path recommendation device, referring to... Figure 6 The diagram illustrates a block diagram of a dynamic path recommendation device according to an embodiment of the present invention. The device includes modules 601-611.

[0057] The historical data acquisition module 601 can be configured to acquire the vehicle's historical driving trajectory and the user's historical navigation records.

[0058] The trajectory data extraction module 602 can be configured to extract the time and location of trajectory data points from the vehicle's historical driving trajectory.

[0059] The time-domain vector generation module 603 can be configured to generate time-domain vectors of points of interest based on the time and location of trajectory data points.

[0060] The navigation data extraction module 604 can be configured to extract the destination type, destination location, and arrival time for each record in the user's historical navigation records.

[0061] The airspace vector generation module 605 can be configured to generate point-of-interest airspace vectors based on the destination type, destination location, and arrival time extracted from the user's historical navigation records.

[0062] The matching module 606 can be configured to match each point of interest in the spatial vector with points in the temporal vector that are at the same or adjacent positions, in order to obtain the length of time spent at each point of interest.

[0063] The spatiotemporal feature map generation module 607 can be configured to generate a spatiotemporal feature map of user behavior habits based on the length of time spent at each point of interest, the location of the point of interest, and the type of the point of interest.

[0064] The format conversion module 608 can be configured to convert spatiotemporal feature maps of user behavior habits into graph data formats.

[0065] The spatiotemporal convolution module 609 can be configured to perform multiple spatiotemporal convolution processes on the spatiotemporal feature map of user behavior habits in graph data format through a multi-layer spatiotemporal convolutional network. Each layer of the multi-layer spatiotemporal convolutional network includes a temporal graph convolutional network and a spatial graph convolutional network. The features obtained by the previous spatiotemporal convolutional network are normalized and then input into the next spatiotemporal convolutional network.

[0066] The heatmap acquisition module 610 can be configured to normalize and upsample the features obtained through a multi-layer spatiotemporal convolutional network to obtain a heatmap of user behavior habits.

[0067] The route determination module 611 can be configured to determine a recommended route based on the vehicle's current location, real-time traffic conditions, and a heatmap of user behavior habits.

[0068] As one embodiment of the present invention, the device may further include an attention mechanism module configured to: after generating a spatiotemporal feature map of user behavior habits, assign weights to each interest point according to the priority of each interest point through an attention mechanism; and update the spatiotemporal feature map of user behavior habits according to the weights assigned to each interest point.

[0069] As one embodiment of the present invention, the device may further include a weight update module configured to: receive user operations on the recommended path; and update the weights of points of interest that are the destinations of the recommended path according to the user operations.

[0070] As one embodiment of the present invention, the spatiotemporal convolution module can be further configured to: input the unnormalized features output by the first spatiotemporal convolutional network in the multi-layer spatiotemporal convolutional network into each subsequent spatiotemporal convolutional network.

[0071] It should be noted that the functions implemented by each module in the dynamic path recommendation device proposed in the embodiments of the present invention correspond one-to-one with each step of the dynamic path recommendation method described above. For other specific embodiments, examples and beneficial effects of the device, please refer to the description of the method above.

[0072] In another aspect, embodiments of the present invention provide a vehicle infotainment system including a memory and a processor, the memory storing a computer program that, when executed by the processor, provides the dynamic path recommendation method described in any of the above embodiments.

[0073] In another aspect, embodiments of the present invention provide a storage medium storing computer-readable instructions that, when executed by a processor, perform the dynamic path recommendation method described in any of the above embodiments.

[0074] The embodiments of the present invention provide a dynamic route guidance scheme. This scheme not only recommends routes based on the real-time traffic conditions of the city, but also guides users to suitable points of interest to engage in corresponding activities when congestion occurs, based on the user's behavioral habits, while waiting for the congestion to ease or end.

[0075] The technical solution adopted by the embodiments of the present invention to solve its technical problem is as follows: a heat map of urban points of interest is established based on user behavior habits. In order to establish a heat map of points of interest, it is first necessary to obtain vehicle driving trajectory and user navigation record. Then, the two types of data are fused to obtain a spatiotemporal feature map of user behavior habits. Next, a graph neural network model is established based on the data. Finally, the spatiotemporal feature map of user behavior habits is fed into the model for training to obtain a heat map of urban points of interest used to guide users.

[0076] Whenever a user encounters unavoidable traffic congestion while driving, the vehicle's infotainment system will recommend congestion avoidance options based on the user's city interest heatmap, guiding the user to points of interest that match their behavioral habits and preventing the user from getting stuck in the vast traffic.

[0077] In addition, by periodically asking users whether they want to update the model and obtaining their consent, the system uploads recent spatiotemporal feature maps of user behavior habits to the cloud for model fine-tuning and online model upgrades, providing users with a better user experience and thus meeting their ever-changing needs.

[0078] The foregoing description of embodiments of the invention has been given for illustrative purposes and is not exhaustive, nor is it intended to limit the invention to the exact forms disclosed. Those skilled in the art will understand that various changes can be made without departing from the scope of the invention, and elements therein can be substituted with equivalents. Furthermore, many modifications can be made to adapt particular situations or materials to the teachings of the invention without departing from the basic scope of the invention. Therefore, the invention is not intended to be limited to the specific embodiments disclosed as the best mode contemplated for carrying out the invention; the invention will include all embodiments falling within the scope of the appended claims.

Claims

1. A dynamic path recommendation method, characterized by, The method comprises: acquiring a vehicle historical driving track and a user historical navigation record; extracting time and position of a track data point in the vehicle historical driving track; generating a time-domain vector of a point of interest according to the time and position of the track data point; extracting a destination type, a destination position and an arrival time of each record in the user historical navigation record; generating a space-domain vector of a point of interest according to the extracted destination type, destination position and arrival time in the user historical navigation record; matching each point of interest in the space-domain vector of a point of interest with a point of interest in the time-domain vector of a point of interest with the same position or adjacent position to obtain a length of time of staying at each point of interest; generating a space-time feature map of a user behavior habit according to the length of time of staying at each point of interest, a position of a point of interest and a type of a point of interest; converting the space-time feature map of a user behavior habit into a graph data format; performing multiple times of space-time convolution processing on the space-time feature map of a user behavior habit in the graph data format by a multi-layer space-time convolution network, wherein each layer of the multi-layer space-time convolution network comprises a time-series graph convolution network and a space graph convolution network, and wherein a feature obtained by a previous layer of space-time convolution network is input into a next layer of space-time convolution network after normalization processing; performing normalization processing and up-sampling on a feature obtained by the multi-layer space-time convolution network to obtain a user behavior habit heat map; determining a recommended path according to a current position of a vehicle, a current real-time road condition and the user behavior habit heat map.

2. The method of claim 1, wherein, After generating the space-time feature map of a user behavior habit, the method further comprises: allocating a weight to each point of interest by an attention mechanism according to a priority of each point of interest; updating the space-time feature map of a user behavior habit according to the weight allocated to each point of interest.

3. The method of claim 2, wherein, The method further comprises: receiving an operation of a user on the recommended path; updating a weight of a point of interest as a destination of the recommended path according to the operation of the user.

4. The method of claim 1, wherein, Performing multiple times of space-time convolution processing comprises: inputting a feature output by a first layer of space-time convolution network in the multi-layer space-time convolution network into each subsequent layer of space-time convolution network without normalization processing.

5. The method of claim 1, wherein, Performing multiple times of space-time convolution processing comprises: in each layer of space-time convolution network, first processing by a time-series graph convolution network and then processing by a space graph convolution network.

6. The method of claim 1, wherein, The method further comprises: after all points of interest in the space-domain vector of a point of interest have been matched, if there are still points of interest with long-term staying and not matched in the time-domain vector of a point of interest, extracting a position, an arrival time and a staying time of these points of interest; determining a type of a point of interest according to the position of the point of interest; filling the extracted position, arrival time, staying time and type of the point of interest into the space-time feature map of a user behavior habit.

7. The method of claim 1, wherein, The multi-layer space-time convolution network comprises three layers of space-time convolution network.

8. The method of claim 1, wherein, The method further comprises: sending a prompt message of whether to update a model through a human-computer interface; receiving a selection of the user on the prompt message; in a case where the user selects to update the model, performing model upgrading on the multi-layer space-time convolution network by using a recent space-time feature map of a user behavior habit.

9. A dynamic route recommendation apparatus, characterized by comprising: The device comprises: a historical data acquisition module configured to acquire a vehicle historical driving trajectory and a user historical navigation record; a trajectory data extraction module configured to extract time and position of a trajectory data point in the vehicle historical driving trajectory; a time-domain vector generation module configured to generate a point of interest time-domain vector according to the time and position of the trajectory data point; a navigation data extraction module configured to extract a destination type, a destination position and an arrival time of each record in the user historical navigation record; a space-domain vector generation module configured to generate a point of interest space-domain vector according to the extracted destination type, destination position and arrival time in the user historical navigation record; a matching module configured to match each point of interest in the point of interest space-domain vector with a point of interest in the point of interest time-domain vector that has the same or adjacent position, to obtain a length of time of staying at each point of interest; a space-time feature map generation module configured to generate a user behavior habit space-time feature map according to the length of time of staying at each point of interest, a point of interest position and a point of interest type; a format conversion module configured to convert the user behavior habit space-time feature map into a graph data format; a space-time convolution module configured to sequentially perform multiple space-time convolution processing on the user behavior habit space-time feature map in the graph data format through a multi-layer space-time convolution network, wherein each layer of the multi-layer space-time convolution network includes a time series graph convolution network and a spatial graph convolution network, and wherein features obtained by a previous layer of space-time convolution network processing are input into a next layer of space-time convolution network after normalization processing; a heat map acquisition module configured to perform normalization processing and up-sampling on features obtained by the multi-layer space-time convolution network, to obtain a user behavior habit heat map; a path determination module configured to determine a recommended path according to a current position of a vehicle, a current real-time road condition and the user behavior habit heat map.

10. The apparatus of claim 9, wherein, The device further includes an attention mechanism module configured to: after generating the user behavior habit space-time feature map, assign a weight to each point of interest through an attention mechanism according to a priority of each point of interest; and update the user behavior habit space-time feature map according to the weight assigned to each point of interest.

11. The apparatus of claim 10, wherein, The device further includes a weight update module configured to: receive an operation of a user on the recommended path; and update a weight of a point of interest that is a destination of the recommended path according to the operation of the user.

12. The apparatus of claim 9, wherein, The space-time convolution module is further configured to: input features output by a first layer of space-time convolution network in the multi-layer space-time convolution network into each subsequent layer of space-time convolution network without normalization processing.

13. A head unit system characterized by comprising: The in-vehicle infotainment system includes a memory and a processor, and the memory stores a computer program that, when executed by the processor, implements the dynamic path recommendation method according to any one of claims 1-8.

14. A storage medium storing computer-readable instructions that, when executed by a processor, perform the dynamic path recommendation method according to any one of claims 1-8.

Images